sbp2: update 36byte inquiry workaround (fix compatibility regression)
[deliverable/linux.git] / net / xfrm / xfrm_policy.c
CommitLineData
1da177e4
LT
1/*
2 * xfrm_policy.c
3 *
4 * Changes:
5 * Mitsuru KANDA @USAGI
6 * Kazunori MIYAZAWA @USAGI
7 * Kunihiro Ishiguro <kunihiro@ipinfusion.com>
8 * IPv6 support
9 * Kazunori MIYAZAWA @USAGI
10 * YOSHIFUJI Hideaki
11 * Split up af-specific portion
12 * Derek Atkins <derek@ihtfp.com> Add the post_input processor
df71837d 13 *
1da177e4
LT
14 */
15
1da177e4
LT
16#include <linux/config.h>
17#include <linux/slab.h>
18#include <linux/kmod.h>
19#include <linux/list.h>
20#include <linux/spinlock.h>
21#include <linux/workqueue.h>
22#include <linux/notifier.h>
23#include <linux/netdevice.h>
eb9c7ebe 24#include <linux/netfilter.h>
1da177e4
LT
25#include <linux/module.h>
26#include <net/xfrm.h>
27#include <net/ip.h>
28
29DECLARE_MUTEX(xfrm_cfg_sem);
30EXPORT_SYMBOL(xfrm_cfg_sem);
31
32static DEFINE_RWLOCK(xfrm_policy_lock);
33
34struct xfrm_policy *xfrm_policy_list[XFRM_POLICY_MAX*2];
35EXPORT_SYMBOL(xfrm_policy_list);
36
37static DEFINE_RWLOCK(xfrm_policy_afinfo_lock);
38static struct xfrm_policy_afinfo *xfrm_policy_afinfo[NPROTO];
39
ba89966c 40static kmem_cache_t *xfrm_dst_cache __read_mostly;
1da177e4
LT
41
42static struct work_struct xfrm_policy_gc_work;
43static struct list_head xfrm_policy_gc_list =
44 LIST_HEAD_INIT(xfrm_policy_gc_list);
45static DEFINE_SPINLOCK(xfrm_policy_gc_lock);
46
47static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family);
48static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo);
49
50int xfrm_register_type(struct xfrm_type *type, unsigned short family)
51{
52 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
53 struct xfrm_type_map *typemap;
54 int err = 0;
55
56 if (unlikely(afinfo == NULL))
57 return -EAFNOSUPPORT;
58 typemap = afinfo->type_map;
59
60 write_lock(&typemap->lock);
61 if (likely(typemap->map[type->proto] == NULL))
62 typemap->map[type->proto] = type;
63 else
64 err = -EEXIST;
65 write_unlock(&typemap->lock);
66 xfrm_policy_put_afinfo(afinfo);
67 return err;
68}
69EXPORT_SYMBOL(xfrm_register_type);
70
71int xfrm_unregister_type(struct xfrm_type *type, unsigned short family)
72{
73 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
74 struct xfrm_type_map *typemap;
75 int err = 0;
76
77 if (unlikely(afinfo == NULL))
78 return -EAFNOSUPPORT;
79 typemap = afinfo->type_map;
80
81 write_lock(&typemap->lock);
82 if (unlikely(typemap->map[type->proto] != type))
83 err = -ENOENT;
84 else
85 typemap->map[type->proto] = NULL;
86 write_unlock(&typemap->lock);
87 xfrm_policy_put_afinfo(afinfo);
88 return err;
89}
90EXPORT_SYMBOL(xfrm_unregister_type);
91
92struct xfrm_type *xfrm_get_type(u8 proto, unsigned short family)
93{
94 struct xfrm_policy_afinfo *afinfo;
95 struct xfrm_type_map *typemap;
96 struct xfrm_type *type;
97 int modload_attempted = 0;
98
99retry:
100 afinfo = xfrm_policy_get_afinfo(family);
101 if (unlikely(afinfo == NULL))
102 return NULL;
103 typemap = afinfo->type_map;
104
105 read_lock(&typemap->lock);
106 type = typemap->map[proto];
107 if (unlikely(type && !try_module_get(type->owner)))
108 type = NULL;
109 read_unlock(&typemap->lock);
110 if (!type && !modload_attempted) {
111 xfrm_policy_put_afinfo(afinfo);
112 request_module("xfrm-type-%d-%d",
113 (int) family, (int) proto);
114 modload_attempted = 1;
115 goto retry;
116 }
117
118 xfrm_policy_put_afinfo(afinfo);
119 return type;
120}
1da177e4
LT
121
122int xfrm_dst_lookup(struct xfrm_dst **dst, struct flowi *fl,
123 unsigned short family)
124{
125 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
126 int err = 0;
127
128 if (unlikely(afinfo == NULL))
129 return -EAFNOSUPPORT;
130
131 if (likely(afinfo->dst_lookup != NULL))
132 err = afinfo->dst_lookup(dst, fl);
133 else
134 err = -EINVAL;
135 xfrm_policy_put_afinfo(afinfo);
136 return err;
137}
138EXPORT_SYMBOL(xfrm_dst_lookup);
139
140void xfrm_put_type(struct xfrm_type *type)
141{
142 module_put(type->owner);
143}
144
145static inline unsigned long make_jiffies(long secs)
146{
147 if (secs >= (MAX_SCHEDULE_TIMEOUT-1)/HZ)
148 return MAX_SCHEDULE_TIMEOUT-1;
149 else
150 return secs*HZ;
151}
152
153static void xfrm_policy_timer(unsigned long data)
154{
155 struct xfrm_policy *xp = (struct xfrm_policy*)data;
156 unsigned long now = (unsigned long)xtime.tv_sec;
157 long next = LONG_MAX;
158 int warn = 0;
159 int dir;
160
161 read_lock(&xp->lock);
162
163 if (xp->dead)
164 goto out;
165
77d8d7a6 166 dir = xfrm_policy_id2dir(xp->index);
1da177e4
LT
167
168 if (xp->lft.hard_add_expires_seconds) {
169 long tmo = xp->lft.hard_add_expires_seconds +
170 xp->curlft.add_time - now;
171 if (tmo <= 0)
172 goto expired;
173 if (tmo < next)
174 next = tmo;
175 }
176 if (xp->lft.hard_use_expires_seconds) {
177 long tmo = xp->lft.hard_use_expires_seconds +
178 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
179 if (tmo <= 0)
180 goto expired;
181 if (tmo < next)
182 next = tmo;
183 }
184 if (xp->lft.soft_add_expires_seconds) {
185 long tmo = xp->lft.soft_add_expires_seconds +
186 xp->curlft.add_time - now;
187 if (tmo <= 0) {
188 warn = 1;
189 tmo = XFRM_KM_TIMEOUT;
190 }
191 if (tmo < next)
192 next = tmo;
193 }
194 if (xp->lft.soft_use_expires_seconds) {
195 long tmo = xp->lft.soft_use_expires_seconds +
196 (xp->curlft.use_time ? : xp->curlft.add_time) - now;
197 if (tmo <= 0) {
198 warn = 1;
199 tmo = XFRM_KM_TIMEOUT;
200 }
201 if (tmo < next)
202 next = tmo;
203 }
204
205 if (warn)
206 km_policy_expired(xp, dir, 0);
207 if (next != LONG_MAX &&
208 !mod_timer(&xp->timer, jiffies + make_jiffies(next)))
209 xfrm_pol_hold(xp);
210
211out:
212 read_unlock(&xp->lock);
213 xfrm_pol_put(xp);
214 return;
215
216expired:
217 read_unlock(&xp->lock);
4666faab
HX
218 if (!xfrm_policy_delete(xp, dir))
219 km_policy_expired(xp, dir, 1);
1da177e4
LT
220 xfrm_pol_put(xp);
221}
222
223
224/* Allocate xfrm_policy. Not used here, it is supposed to be used by pfkeyv2
225 * SPD calls.
226 */
227
dd0fc66f 228struct xfrm_policy *xfrm_policy_alloc(gfp_t gfp)
1da177e4
LT
229{
230 struct xfrm_policy *policy;
231
232 policy = kmalloc(sizeof(struct xfrm_policy), gfp);
233
234 if (policy) {
235 memset(policy, 0, sizeof(struct xfrm_policy));
236 atomic_set(&policy->refcnt, 1);
237 rwlock_init(&policy->lock);
238 init_timer(&policy->timer);
239 policy->timer.data = (unsigned long)policy;
240 policy->timer.function = xfrm_policy_timer;
241 }
242 return policy;
243}
244EXPORT_SYMBOL(xfrm_policy_alloc);
245
246/* Destroy xfrm_policy: descendant resources must be released to this moment. */
247
248void __xfrm_policy_destroy(struct xfrm_policy *policy)
249{
09a62660 250 BUG_ON(!policy->dead);
1da177e4 251
09a62660 252 BUG_ON(policy->bundles);
1da177e4
LT
253
254 if (del_timer(&policy->timer))
255 BUG();
256
df71837d 257 security_xfrm_policy_free(policy);
1da177e4
LT
258 kfree(policy);
259}
260EXPORT_SYMBOL(__xfrm_policy_destroy);
261
262static void xfrm_policy_gc_kill(struct xfrm_policy *policy)
263{
264 struct dst_entry *dst;
265
266 while ((dst = policy->bundles) != NULL) {
267 policy->bundles = dst->next;
268 dst_free(dst);
269 }
270
271 if (del_timer(&policy->timer))
272 atomic_dec(&policy->refcnt);
273
274 if (atomic_read(&policy->refcnt) > 1)
275 flow_cache_flush();
276
277 xfrm_pol_put(policy);
278}
279
280static void xfrm_policy_gc_task(void *data)
281{
282 struct xfrm_policy *policy;
283 struct list_head *entry, *tmp;
284 struct list_head gc_list = LIST_HEAD_INIT(gc_list);
285
286 spin_lock_bh(&xfrm_policy_gc_lock);
287 list_splice_init(&xfrm_policy_gc_list, &gc_list);
288 spin_unlock_bh(&xfrm_policy_gc_lock);
289
290 list_for_each_safe(entry, tmp, &gc_list) {
291 policy = list_entry(entry, struct xfrm_policy, list);
292 xfrm_policy_gc_kill(policy);
293 }
294}
295
296/* Rule must be locked. Release descentant resources, announce
297 * entry dead. The rule must be unlinked from lists to the moment.
298 */
299
300static void xfrm_policy_kill(struct xfrm_policy *policy)
301{
302 int dead;
303
304 write_lock_bh(&policy->lock);
305 dead = policy->dead;
306 policy->dead = 1;
307 write_unlock_bh(&policy->lock);
308
309 if (unlikely(dead)) {
310 WARN_ON(1);
311 return;
312 }
313
314 spin_lock(&xfrm_policy_gc_lock);
315 list_add(&policy->list, &xfrm_policy_gc_list);
316 spin_unlock(&xfrm_policy_gc_lock);
317
318 schedule_work(&xfrm_policy_gc_work);
319}
320
321/* Generate new index... KAME seems to generate them ordered by cost
322 * of an absolute inpredictability of ordering of rules. This will not pass. */
323static u32 xfrm_gen_index(int dir)
324{
325 u32 idx;
326 struct xfrm_policy *p;
327 static u32 idx_generator;
328
329 for (;;) {
330 idx = (idx_generator | dir);
331 idx_generator += 8;
332 if (idx == 0)
333 idx = 8;
334 for (p = xfrm_policy_list[dir]; p; p = p->next) {
335 if (p->index == idx)
336 break;
337 }
338 if (!p)
339 return idx;
340 }
341}
342
343int xfrm_policy_insert(int dir, struct xfrm_policy *policy, int excl)
344{
345 struct xfrm_policy *pol, **p;
346 struct xfrm_policy *delpol = NULL;
347 struct xfrm_policy **newpos = NULL;
9b78a82c 348 struct dst_entry *gc_list;
1da177e4
LT
349
350 write_lock_bh(&xfrm_policy_lock);
351 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL;) {
df71837d
TJ
352 if (!delpol && memcmp(&policy->selector, &pol->selector, sizeof(pol->selector)) == 0 &&
353 xfrm_sec_ctx_match(pol->security, policy->security)) {
1da177e4
LT
354 if (excl) {
355 write_unlock_bh(&xfrm_policy_lock);
356 return -EEXIST;
357 }
358 *p = pol->next;
359 delpol = pol;
360 if (policy->priority > pol->priority)
361 continue;
362 } else if (policy->priority >= pol->priority) {
363 p = &pol->next;
364 continue;
365 }
366 if (!newpos)
367 newpos = p;
368 if (delpol)
369 break;
370 p = &pol->next;
371 }
372 if (newpos)
373 p = newpos;
374 xfrm_pol_hold(policy);
375 policy->next = *p;
376 *p = policy;
377 atomic_inc(&flow_cache_genid);
378 policy->index = delpol ? delpol->index : xfrm_gen_index(dir);
379 policy->curlft.add_time = (unsigned long)xtime.tv_sec;
380 policy->curlft.use_time = 0;
381 if (!mod_timer(&policy->timer, jiffies + HZ))
382 xfrm_pol_hold(policy);
383 write_unlock_bh(&xfrm_policy_lock);
384
9b78a82c 385 if (delpol)
1da177e4 386 xfrm_policy_kill(delpol);
9b78a82c
DM
387
388 read_lock_bh(&xfrm_policy_lock);
389 gc_list = NULL;
390 for (policy = policy->next; policy; policy = policy->next) {
391 struct dst_entry *dst;
392
393 write_lock(&policy->lock);
394 dst = policy->bundles;
395 if (dst) {
396 struct dst_entry *tail = dst;
397 while (tail->next)
398 tail = tail->next;
399 tail->next = gc_list;
400 gc_list = dst;
401
402 policy->bundles = NULL;
403 }
404 write_unlock(&policy->lock);
1da177e4 405 }
9b78a82c
DM
406 read_unlock_bh(&xfrm_policy_lock);
407
408 while (gc_list) {
409 struct dst_entry *dst = gc_list;
410
411 gc_list = dst->next;
412 dst_free(dst);
413 }
414
1da177e4
LT
415 return 0;
416}
417EXPORT_SYMBOL(xfrm_policy_insert);
418
df71837d
TJ
419struct xfrm_policy *xfrm_policy_bysel_ctx(int dir, struct xfrm_selector *sel,
420 struct xfrm_sec_ctx *ctx, int delete)
1da177e4
LT
421{
422 struct xfrm_policy *pol, **p;
423
424 write_lock_bh(&xfrm_policy_lock);
425 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
df71837d
TJ
426 if ((memcmp(sel, &pol->selector, sizeof(*sel)) == 0) &&
427 (xfrm_sec_ctx_match(ctx, pol->security))) {
1da177e4
LT
428 xfrm_pol_hold(pol);
429 if (delete)
430 *p = pol->next;
431 break;
432 }
433 }
434 write_unlock_bh(&xfrm_policy_lock);
435
436 if (pol && delete) {
437 atomic_inc(&flow_cache_genid);
438 xfrm_policy_kill(pol);
439 }
440 return pol;
441}
df71837d 442EXPORT_SYMBOL(xfrm_policy_bysel_ctx);
1da177e4
LT
443
444struct xfrm_policy *xfrm_policy_byid(int dir, u32 id, int delete)
445{
446 struct xfrm_policy *pol, **p;
447
448 write_lock_bh(&xfrm_policy_lock);
77d8d7a6 449 for (p = &xfrm_policy_list[dir]; (pol=*p)!=NULL; p = &pol->next) {
1da177e4
LT
450 if (pol->index == id) {
451 xfrm_pol_hold(pol);
452 if (delete)
453 *p = pol->next;
454 break;
455 }
456 }
457 write_unlock_bh(&xfrm_policy_lock);
458
459 if (pol && delete) {
460 atomic_inc(&flow_cache_genid);
461 xfrm_policy_kill(pol);
462 }
463 return pol;
464}
465EXPORT_SYMBOL(xfrm_policy_byid);
466
467void xfrm_policy_flush(void)
468{
469 struct xfrm_policy *xp;
470 int dir;
471
472 write_lock_bh(&xfrm_policy_lock);
473 for (dir = 0; dir < XFRM_POLICY_MAX; dir++) {
474 while ((xp = xfrm_policy_list[dir]) != NULL) {
475 xfrm_policy_list[dir] = xp->next;
476 write_unlock_bh(&xfrm_policy_lock);
477
478 xfrm_policy_kill(xp);
479
480 write_lock_bh(&xfrm_policy_lock);
481 }
482 }
483 atomic_inc(&flow_cache_genid);
484 write_unlock_bh(&xfrm_policy_lock);
485}
486EXPORT_SYMBOL(xfrm_policy_flush);
487
488int xfrm_policy_walk(int (*func)(struct xfrm_policy *, int, int, void*),
489 void *data)
490{
491 struct xfrm_policy *xp;
492 int dir;
493 int count = 0;
494 int error = 0;
495
496 read_lock_bh(&xfrm_policy_lock);
497 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
498 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next)
499 count++;
500 }
501
502 if (count == 0) {
503 error = -ENOENT;
504 goto out;
505 }
506
507 for (dir = 0; dir < 2*XFRM_POLICY_MAX; dir++) {
508 for (xp = xfrm_policy_list[dir]; xp; xp = xp->next) {
509 error = func(xp, dir%XFRM_POLICY_MAX, --count, data);
510 if (error)
511 goto out;
512 }
513 }
514
515out:
516 read_unlock_bh(&xfrm_policy_lock);
517 return error;
518}
519EXPORT_SYMBOL(xfrm_policy_walk);
520
521/* Find policy to apply to this flow. */
522
df71837d 523static void xfrm_policy_lookup(struct flowi *fl, u32 sk_sid, u16 family, u8 dir,
1da177e4
LT
524 void **objp, atomic_t **obj_refp)
525{
526 struct xfrm_policy *pol;
527
528 read_lock_bh(&xfrm_policy_lock);
529 for (pol = xfrm_policy_list[dir]; pol; pol = pol->next) {
530 struct xfrm_selector *sel = &pol->selector;
531 int match;
532
533 if (pol->family != family)
534 continue;
535
536 match = xfrm_selector_match(sel, fl, family);
df71837d 537
1da177e4 538 if (match) {
df71837d
TJ
539 if (!security_xfrm_policy_lookup(pol, sk_sid, dir)) {
540 xfrm_pol_hold(pol);
541 break;
542 }
1da177e4
LT
543 }
544 }
545 read_unlock_bh(&xfrm_policy_lock);
546 if ((*objp = (void *) pol) != NULL)
547 *obj_refp = &pol->refcnt;
548}
549
df71837d
TJ
550static inline int policy_to_flow_dir(int dir)
551{
552 if (XFRM_POLICY_IN == FLOW_DIR_IN &&
553 XFRM_POLICY_OUT == FLOW_DIR_OUT &&
554 XFRM_POLICY_FWD == FLOW_DIR_FWD)
555 return dir;
556 switch (dir) {
557 default:
558 case XFRM_POLICY_IN:
559 return FLOW_DIR_IN;
560 case XFRM_POLICY_OUT:
561 return FLOW_DIR_OUT;
562 case XFRM_POLICY_FWD:
563 return FLOW_DIR_FWD;
564 };
565}
566
567static struct xfrm_policy *xfrm_sk_policy_lookup(struct sock *sk, int dir, struct flowi *fl, u32 sk_sid)
1da177e4
LT
568{
569 struct xfrm_policy *pol;
570
571 read_lock_bh(&xfrm_policy_lock);
572 if ((pol = sk->sk_policy[dir]) != NULL) {
df71837d 573 int match = xfrm_selector_match(&pol->selector, fl,
1da177e4 574 sk->sk_family);
df71837d
TJ
575 int err = 0;
576
1da177e4 577 if (match)
df71837d
TJ
578 err = security_xfrm_policy_lookup(pol, sk_sid, policy_to_flow_dir(dir));
579
580 if (match && !err)
1da177e4
LT
581 xfrm_pol_hold(pol);
582 else
583 pol = NULL;
584 }
585 read_unlock_bh(&xfrm_policy_lock);
586 return pol;
587}
588
589static void __xfrm_policy_link(struct xfrm_policy *pol, int dir)
590{
591 pol->next = xfrm_policy_list[dir];
592 xfrm_policy_list[dir] = pol;
593 xfrm_pol_hold(pol);
594}
595
596static struct xfrm_policy *__xfrm_policy_unlink(struct xfrm_policy *pol,
597 int dir)
598{
599 struct xfrm_policy **polp;
600
601 for (polp = &xfrm_policy_list[dir];
602 *polp != NULL; polp = &(*polp)->next) {
603 if (*polp == pol) {
604 *polp = pol->next;
605 return pol;
606 }
607 }
608 return NULL;
609}
610
4666faab 611int xfrm_policy_delete(struct xfrm_policy *pol, int dir)
1da177e4
LT
612{
613 write_lock_bh(&xfrm_policy_lock);
614 pol = __xfrm_policy_unlink(pol, dir);
615 write_unlock_bh(&xfrm_policy_lock);
616 if (pol) {
617 if (dir < XFRM_POLICY_MAX)
618 atomic_inc(&flow_cache_genid);
619 xfrm_policy_kill(pol);
4666faab 620 return 0;
1da177e4 621 }
4666faab 622 return -ENOENT;
1da177e4
LT
623}
624
625int xfrm_sk_policy_insert(struct sock *sk, int dir, struct xfrm_policy *pol)
626{
627 struct xfrm_policy *old_pol;
628
629 write_lock_bh(&xfrm_policy_lock);
630 old_pol = sk->sk_policy[dir];
631 sk->sk_policy[dir] = pol;
632 if (pol) {
633 pol->curlft.add_time = (unsigned long)xtime.tv_sec;
634 pol->index = xfrm_gen_index(XFRM_POLICY_MAX+dir);
635 __xfrm_policy_link(pol, XFRM_POLICY_MAX+dir);
636 }
637 if (old_pol)
638 __xfrm_policy_unlink(old_pol, XFRM_POLICY_MAX+dir);
639 write_unlock_bh(&xfrm_policy_lock);
640
641 if (old_pol) {
642 xfrm_policy_kill(old_pol);
643 }
644 return 0;
645}
646
647static struct xfrm_policy *clone_policy(struct xfrm_policy *old, int dir)
648{
649 struct xfrm_policy *newp = xfrm_policy_alloc(GFP_ATOMIC);
650
651 if (newp) {
652 newp->selector = old->selector;
df71837d
TJ
653 if (security_xfrm_policy_clone(old, newp)) {
654 kfree(newp);
655 return NULL; /* ENOMEM */
656 }
1da177e4
LT
657 newp->lft = old->lft;
658 newp->curlft = old->curlft;
659 newp->action = old->action;
660 newp->flags = old->flags;
661 newp->xfrm_nr = old->xfrm_nr;
662 newp->index = old->index;
663 memcpy(newp->xfrm_vec, old->xfrm_vec,
664 newp->xfrm_nr*sizeof(struct xfrm_tmpl));
665 write_lock_bh(&xfrm_policy_lock);
666 __xfrm_policy_link(newp, XFRM_POLICY_MAX+dir);
667 write_unlock_bh(&xfrm_policy_lock);
668 xfrm_pol_put(newp);
669 }
670 return newp;
671}
672
673int __xfrm_sk_clone_policy(struct sock *sk)
674{
675 struct xfrm_policy *p0 = sk->sk_policy[0],
676 *p1 = sk->sk_policy[1];
677
678 sk->sk_policy[0] = sk->sk_policy[1] = NULL;
679 if (p0 && (sk->sk_policy[0] = clone_policy(p0, 0)) == NULL)
680 return -ENOMEM;
681 if (p1 && (sk->sk_policy[1] = clone_policy(p1, 1)) == NULL)
682 return -ENOMEM;
683 return 0;
684}
685
686/* Resolve list of templates for the flow, given policy. */
687
688static int
689xfrm_tmpl_resolve(struct xfrm_policy *policy, struct flowi *fl,
690 struct xfrm_state **xfrm,
691 unsigned short family)
692{
693 int nx;
694 int i, error;
695 xfrm_address_t *daddr = xfrm_flowi_daddr(fl, family);
696 xfrm_address_t *saddr = xfrm_flowi_saddr(fl, family);
697
698 for (nx=0, i = 0; i < policy->xfrm_nr; i++) {
699 struct xfrm_state *x;
700 xfrm_address_t *remote = daddr;
701 xfrm_address_t *local = saddr;
702 struct xfrm_tmpl *tmpl = &policy->xfrm_vec[i];
703
704 if (tmpl->mode) {
705 remote = &tmpl->id.daddr;
706 local = &tmpl->saddr;
707 }
708
709 x = xfrm_state_find(remote, local, fl, tmpl, policy, &error, family);
710
711 if (x && x->km.state == XFRM_STATE_VALID) {
712 xfrm[nx++] = x;
713 daddr = remote;
714 saddr = local;
715 continue;
716 }
717 if (x) {
718 error = (x->km.state == XFRM_STATE_ERROR ?
719 -EINVAL : -EAGAIN);
720 xfrm_state_put(x);
721 }
722
723 if (!tmpl->optional)
724 goto fail;
725 }
726 return nx;
727
728fail:
729 for (nx--; nx>=0; nx--)
730 xfrm_state_put(xfrm[nx]);
731 return error;
732}
733
734/* Check that the bundle accepts the flow and its components are
735 * still valid.
736 */
737
738static struct dst_entry *
739xfrm_find_bundle(struct flowi *fl, struct xfrm_policy *policy, unsigned short family)
740{
741 struct dst_entry *x;
742 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
743 if (unlikely(afinfo == NULL))
744 return ERR_PTR(-EINVAL);
745 x = afinfo->find_bundle(fl, policy);
746 xfrm_policy_put_afinfo(afinfo);
747 return x;
748}
749
750/* Allocate chain of dst_entry's, attach known xfrm's, calculate
751 * all the metrics... Shortly, bundle a bundle.
752 */
753
754static int
755xfrm_bundle_create(struct xfrm_policy *policy, struct xfrm_state **xfrm, int nx,
756 struct flowi *fl, struct dst_entry **dst_p,
757 unsigned short family)
758{
759 int err;
760 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
761 if (unlikely(afinfo == NULL))
762 return -EINVAL;
763 err = afinfo->bundle_create(policy, xfrm, nx, fl, dst_p);
764 xfrm_policy_put_afinfo(afinfo);
765 return err;
766}
767
1da177e4
LT
768
769static int stale_bundle(struct dst_entry *dst);
770
771/* Main function: finds/creates a bundle for given flow.
772 *
773 * At the moment we eat a raw IP route. Mostly to speed up lookups
774 * on interfaces with disabled IPsec.
775 */
776int xfrm_lookup(struct dst_entry **dst_p, struct flowi *fl,
777 struct sock *sk, int flags)
778{
779 struct xfrm_policy *policy;
780 struct xfrm_state *xfrm[XFRM_MAX_DEPTH];
781 struct dst_entry *dst, *dst_orig = *dst_p;
782 int nx = 0;
783 int err;
784 u32 genid;
785 u16 family = dst_orig->ops->family;
df71837d
TJ
786 u8 dir = policy_to_flow_dir(XFRM_POLICY_OUT);
787 u32 sk_sid = security_sk_sid(sk, fl, dir);
1da177e4
LT
788restart:
789 genid = atomic_read(&flow_cache_genid);
790 policy = NULL;
791 if (sk && sk->sk_policy[1])
df71837d 792 policy = xfrm_sk_policy_lookup(sk, XFRM_POLICY_OUT, fl, sk_sid);
1da177e4
LT
793
794 if (!policy) {
795 /* To accelerate a bit... */
796 if ((dst_orig->flags & DST_NOXFRM) || !xfrm_policy_list[XFRM_POLICY_OUT])
797 return 0;
798
df71837d 799 policy = flow_cache_lookup(fl, sk_sid, family, dir,
1da177e4
LT
800 xfrm_policy_lookup);
801 }
802
803 if (!policy)
804 return 0;
805
806 policy->curlft.use_time = (unsigned long)xtime.tv_sec;
807
808 switch (policy->action) {
809 case XFRM_POLICY_BLOCK:
810 /* Prohibit the flow */
e104411b
PM
811 err = -EPERM;
812 goto error;
1da177e4
LT
813
814 case XFRM_POLICY_ALLOW:
815 if (policy->xfrm_nr == 0) {
816 /* Flow passes not transformed. */
817 xfrm_pol_put(policy);
818 return 0;
819 }
820
821 /* Try to find matching bundle.
822 *
823 * LATER: help from flow cache. It is optional, this
824 * is required only for output policy.
825 */
826 dst = xfrm_find_bundle(fl, policy, family);
827 if (IS_ERR(dst)) {
e104411b
PM
828 err = PTR_ERR(dst);
829 goto error;
1da177e4
LT
830 }
831
832 if (dst)
833 break;
834
835 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
836
837 if (unlikely(nx<0)) {
838 err = nx;
839 if (err == -EAGAIN && flags) {
840 DECLARE_WAITQUEUE(wait, current);
841
842 add_wait_queue(&km_waitq, &wait);
843 set_current_state(TASK_INTERRUPTIBLE);
844 schedule();
845 set_current_state(TASK_RUNNING);
846 remove_wait_queue(&km_waitq, &wait);
847
848 nx = xfrm_tmpl_resolve(policy, fl, xfrm, family);
849
850 if (nx == -EAGAIN && signal_pending(current)) {
851 err = -ERESTART;
852 goto error;
853 }
854 if (nx == -EAGAIN ||
855 genid != atomic_read(&flow_cache_genid)) {
856 xfrm_pol_put(policy);
857 goto restart;
858 }
859 err = nx;
860 }
861 if (err < 0)
862 goto error;
863 }
864 if (nx == 0) {
865 /* Flow passes not transformed. */
866 xfrm_pol_put(policy);
867 return 0;
868 }
869
870 dst = dst_orig;
871 err = xfrm_bundle_create(policy, xfrm, nx, fl, &dst, family);
872
873 if (unlikely(err)) {
874 int i;
875 for (i=0; i<nx; i++)
876 xfrm_state_put(xfrm[i]);
877 goto error;
878 }
879
880 write_lock_bh(&policy->lock);
881 if (unlikely(policy->dead || stale_bundle(dst))) {
882 /* Wow! While we worked on resolving, this
883 * policy has gone. Retry. It is not paranoia,
884 * we just cannot enlist new bundle to dead object.
885 * We can't enlist stable bundles either.
886 */
887 write_unlock_bh(&policy->lock);
1da177e4
LT
888 if (dst)
889 dst_free(dst);
00de651d
HX
890
891 err = -EHOSTUNREACH;
892 goto error;
1da177e4
LT
893 }
894 dst->next = policy->bundles;
895 policy->bundles = dst;
896 dst_hold(dst);
897 write_unlock_bh(&policy->lock);
898 }
899 *dst_p = dst;
900 dst_release(dst_orig);
901 xfrm_pol_put(policy);
902 return 0;
903
904error:
905 dst_release(dst_orig);
906 xfrm_pol_put(policy);
907 *dst_p = NULL;
908 return err;
909}
910EXPORT_SYMBOL(xfrm_lookup);
911
912/* When skb is transformed back to its "native" form, we have to
913 * check policy restrictions. At the moment we make this in maximally
914 * stupid way. Shame on me. :-) Of course, connected sockets must
915 * have policy cached at them.
916 */
917
918static inline int
919xfrm_state_ok(struct xfrm_tmpl *tmpl, struct xfrm_state *x,
920 unsigned short family)
921{
922 if (xfrm_state_kern(x))
923 return tmpl->optional && !xfrm_state_addr_cmp(tmpl, x, family);
924 return x->id.proto == tmpl->id.proto &&
925 (x->id.spi == tmpl->id.spi || !tmpl->id.spi) &&
926 (x->props.reqid == tmpl->reqid || !tmpl->reqid) &&
927 x->props.mode == tmpl->mode &&
928 (tmpl->aalgos & (1<<x->props.aalgo)) &&
929 !(x->props.mode && xfrm_state_addr_cmp(tmpl, x, family));
930}
931
932static inline int
933xfrm_policy_ok(struct xfrm_tmpl *tmpl, struct sec_path *sp, int start,
934 unsigned short family)
935{
936 int idx = start;
937
938 if (tmpl->optional) {
939 if (!tmpl->mode)
940 return start;
941 } else
942 start = -1;
943 for (; idx < sp->len; idx++) {
944 if (xfrm_state_ok(tmpl, sp->x[idx].xvec, family))
945 return ++idx;
946 if (sp->x[idx].xvec->props.mode)
947 break;
948 }
949 return start;
950}
951
3e3850e9
PM
952int
953xfrm_decode_session(struct sk_buff *skb, struct flowi *fl, unsigned short family)
1da177e4
LT
954{
955 struct xfrm_policy_afinfo *afinfo = xfrm_policy_get_afinfo(family);
956
957 if (unlikely(afinfo == NULL))
958 return -EAFNOSUPPORT;
959
960 afinfo->decode_session(skb, fl);
961 xfrm_policy_put_afinfo(afinfo);
962 return 0;
963}
3e3850e9 964EXPORT_SYMBOL(xfrm_decode_session);
1da177e4
LT
965
966static inline int secpath_has_tunnel(struct sec_path *sp, int k)
967{
968 for (; k < sp->len; k++) {
969 if (sp->x[k].xvec->props.mode)
970 return 1;
971 }
972
973 return 0;
974}
975
976int __xfrm_policy_check(struct sock *sk, int dir, struct sk_buff *skb,
977 unsigned short family)
978{
979 struct xfrm_policy *pol;
980 struct flowi fl;
df71837d
TJ
981 u8 fl_dir = policy_to_flow_dir(dir);
982 u32 sk_sid;
1da177e4 983
3e3850e9 984 if (xfrm_decode_session(skb, &fl, family) < 0)
1da177e4 985 return 0;
eb9c7ebe 986 nf_nat_decode_session(skb, &fl, family);
1da177e4 987
df71837d
TJ
988 sk_sid = security_sk_sid(sk, &fl, fl_dir);
989
1da177e4
LT
990 /* First, check used SA against their selectors. */
991 if (skb->sp) {
992 int i;
993
994 for (i=skb->sp->len-1; i>=0; i--) {
df71837d 995 struct sec_decap_state *xvec = &(skb->sp->x[i]);
1da177e4
LT
996 if (!xfrm_selector_match(&xvec->xvec->sel, &fl, family))
997 return 0;
998
999 /* If there is a post_input processor, try running it */
1000 if (xvec->xvec->type->post_input &&
1001 (xvec->xvec->type->post_input)(xvec->xvec,
1002 &(xvec->decap),
1003 skb) != 0)
1004 return 0;
1005 }
1006 }
1007
1008 pol = NULL;
1009 if (sk && sk->sk_policy[dir])
df71837d 1010 pol = xfrm_sk_policy_lookup(sk, dir, &fl, sk_sid);
1da177e4
LT
1011
1012 if (!pol)
df71837d 1013 pol = flow_cache_lookup(&fl, sk_sid, family, fl_dir,
1da177e4
LT
1014 xfrm_policy_lookup);
1015
1016 if (!pol)
1017 return !skb->sp || !secpath_has_tunnel(skb->sp, 0);
1018
1019 pol->curlft.use_time = (unsigned long)xtime.tv_sec;
1020
1021 if (pol->action == XFRM_POLICY_ALLOW) {
1022 struct sec_path *sp;
1023 static struct sec_path dummy;
1024 int i, k;
1025
1026 if ((sp = skb->sp) == NULL)
1027 sp = &dummy;
1028
1029 /* For each tunnel xfrm, find the first matching tmpl.
1030 * For each tmpl before that, find corresponding xfrm.
1031 * Order is _important_. Later we will implement
1032 * some barriers, but at the moment barriers
1033 * are implied between each two transformations.
1034 */
1035 for (i = pol->xfrm_nr-1, k = 0; i >= 0; i--) {
1036 k = xfrm_policy_ok(pol->xfrm_vec+i, sp, k, family);
1037 if (k < 0)
1038 goto reject;
1039 }
1040
1041 if (secpath_has_tunnel(sp, k))
1042 goto reject;
1043
1044 xfrm_pol_put(pol);
1045 return 1;
1046 }
1047
1048reject:
1049 xfrm_pol_put(pol);
1050 return 0;
1051}
1052EXPORT_SYMBOL(__xfrm_policy_check);
1053
1054int __xfrm_route_forward(struct sk_buff *skb, unsigned short family)
1055{
1056 struct flowi fl;
1057
3e3850e9 1058 if (xfrm_decode_session(skb, &fl, family) < 0)
1da177e4
LT
1059 return 0;
1060
1061 return xfrm_lookup(&skb->dst, &fl, NULL, 0) == 0;
1062}
1063EXPORT_SYMBOL(__xfrm_route_forward);
1064
1da177e4
LT
1065static struct dst_entry *xfrm_dst_check(struct dst_entry *dst, u32 cookie)
1066{
399c180a
DM
1067 /* If it is marked obsolete, which is how we even get here,
1068 * then we have purged it from the policy bundle list and we
1069 * did that for a good reason.
1070 */
1da177e4
LT
1071 return NULL;
1072}
1073
1074static int stale_bundle(struct dst_entry *dst)
1075{
1076 return !xfrm_bundle_ok((struct xfrm_dst *)dst, NULL, AF_UNSPEC);
1077}
1078
aabc9761 1079void xfrm_dst_ifdown(struct dst_entry *dst, struct net_device *dev)
1da177e4 1080{
1da177e4
LT
1081 while ((dst = dst->child) && dst->xfrm && dst->dev == dev) {
1082 dst->dev = &loopback_dev;
1083 dev_hold(&loopback_dev);
1084 dev_put(dev);
1085 }
1086}
aabc9761 1087EXPORT_SYMBOL(xfrm_dst_ifdown);
1da177e4
LT
1088
1089static void xfrm_link_failure(struct sk_buff *skb)
1090{
1091 /* Impossible. Such dst must be popped before reaches point of failure. */
1092 return;
1093}
1094
1095static struct dst_entry *xfrm_negative_advice(struct dst_entry *dst)
1096{
1097 if (dst) {
1098 if (dst->obsolete) {
1099 dst_release(dst);
1100 dst = NULL;
1101 }
1102 }
1103 return dst;
1104}
1105
1106static void xfrm_prune_bundles(int (*func)(struct dst_entry *))
1107{
1108 int i;
1109 struct xfrm_policy *pol;
1110 struct dst_entry *dst, **dstp, *gc_list = NULL;
1111
1112 read_lock_bh(&xfrm_policy_lock);
1113 for (i=0; i<2*XFRM_POLICY_MAX; i++) {
1114 for (pol = xfrm_policy_list[i]; pol; pol = pol->next) {
1115 write_lock(&pol->lock);
1116 dstp = &pol->bundles;
1117 while ((dst=*dstp) != NULL) {
1118 if (func(dst)) {
1119 *dstp = dst->next;
1120 dst->next = gc_list;
1121 gc_list = dst;
1122 } else {
1123 dstp = &dst->next;
1124 }
1125 }
1126 write_unlock(&pol->lock);
1127 }
1128 }
1129 read_unlock_bh(&xfrm_policy_lock);
1130
1131 while (gc_list) {
1132 dst = gc_list;
1133 gc_list = dst->next;
1134 dst_free(dst);
1135 }
1136}
1137
1138static int unused_bundle(struct dst_entry *dst)
1139{
1140 return !atomic_read(&dst->__refcnt);
1141}
1142
1143static void __xfrm_garbage_collect(void)
1144{
1145 xfrm_prune_bundles(unused_bundle);
1146}
1147
1148int xfrm_flush_bundles(void)
1149{
1150 xfrm_prune_bundles(stale_bundle);
1151 return 0;
1152}
1153
399c180a
DM
1154static int always_true(struct dst_entry *dst)
1155{
1156 return 1;
1157}
1158
1159void xfrm_flush_all_bundles(void)
1160{
1161 xfrm_prune_bundles(always_true);
1162}
1163
1da177e4
LT
1164void xfrm_init_pmtu(struct dst_entry *dst)
1165{
1166 do {
1167 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1168 u32 pmtu, route_mtu_cached;
1169
1170 pmtu = dst_mtu(dst->child);
1171 xdst->child_mtu_cached = pmtu;
1172
1173 pmtu = xfrm_state_mtu(dst->xfrm, pmtu);
1174
1175 route_mtu_cached = dst_mtu(xdst->route);
1176 xdst->route_mtu_cached = route_mtu_cached;
1177
1178 if (pmtu > route_mtu_cached)
1179 pmtu = route_mtu_cached;
1180
1181 dst->metrics[RTAX_MTU-1] = pmtu;
1182 } while ((dst = dst->next));
1183}
1184
1185EXPORT_SYMBOL(xfrm_init_pmtu);
1186
1187/* Check that the bundle accepts the flow and its components are
1188 * still valid.
1189 */
1190
1191int xfrm_bundle_ok(struct xfrm_dst *first, struct flowi *fl, int family)
1192{
1193 struct dst_entry *dst = &first->u.dst;
1194 struct xfrm_dst *last;
1195 u32 mtu;
1196
92d63dec 1197 if (!dst_check(dst->path, ((struct xfrm_dst *)dst)->path_cookie) ||
1da177e4
LT
1198 (dst->dev && !netif_running(dst->dev)))
1199 return 0;
1200
1201 last = NULL;
1202
1203 do {
1204 struct xfrm_dst *xdst = (struct xfrm_dst *)dst;
1205
1206 if (fl && !xfrm_selector_match(&dst->xfrm->sel, fl, family))
1207 return 0;
1208 if (dst->xfrm->km.state != XFRM_STATE_VALID)
1209 return 0;
1210
1211 mtu = dst_mtu(dst->child);
1212 if (xdst->child_mtu_cached != mtu) {
1213 last = xdst;
1214 xdst->child_mtu_cached = mtu;
1215 }
1216
92d63dec 1217 if (!dst_check(xdst->route, xdst->route_cookie))
1da177e4
LT
1218 return 0;
1219 mtu = dst_mtu(xdst->route);
1220 if (xdst->route_mtu_cached != mtu) {
1221 last = xdst;
1222 xdst->route_mtu_cached = mtu;
1223 }
1224
1225 dst = dst->child;
1226 } while (dst->xfrm);
1227
1228 if (likely(!last))
1229 return 1;
1230
1231 mtu = last->child_mtu_cached;
1232 for (;;) {
1233 dst = &last->u.dst;
1234
1235 mtu = xfrm_state_mtu(dst->xfrm, mtu);
1236 if (mtu > last->route_mtu_cached)
1237 mtu = last->route_mtu_cached;
1238 dst->metrics[RTAX_MTU-1] = mtu;
1239
1240 if (last == first)
1241 break;
1242
1243 last = last->u.next;
1244 last->child_mtu_cached = mtu;
1245 }
1246
1247 return 1;
1248}
1249
1250EXPORT_SYMBOL(xfrm_bundle_ok);
1251
1da177e4
LT
1252int xfrm_policy_register_afinfo(struct xfrm_policy_afinfo *afinfo)
1253{
1254 int err = 0;
1255 if (unlikely(afinfo == NULL))
1256 return -EINVAL;
1257 if (unlikely(afinfo->family >= NPROTO))
1258 return -EAFNOSUPPORT;
1259 write_lock(&xfrm_policy_afinfo_lock);
1260 if (unlikely(xfrm_policy_afinfo[afinfo->family] != NULL))
1261 err = -ENOBUFS;
1262 else {
1263 struct dst_ops *dst_ops = afinfo->dst_ops;
1264 if (likely(dst_ops->kmem_cachep == NULL))
1265 dst_ops->kmem_cachep = xfrm_dst_cache;
1266 if (likely(dst_ops->check == NULL))
1267 dst_ops->check = xfrm_dst_check;
1da177e4
LT
1268 if (likely(dst_ops->negative_advice == NULL))
1269 dst_ops->negative_advice = xfrm_negative_advice;
1270 if (likely(dst_ops->link_failure == NULL))
1271 dst_ops->link_failure = xfrm_link_failure;
1da177e4
LT
1272 if (likely(afinfo->garbage_collect == NULL))
1273 afinfo->garbage_collect = __xfrm_garbage_collect;
1274 xfrm_policy_afinfo[afinfo->family] = afinfo;
1275 }
1276 write_unlock(&xfrm_policy_afinfo_lock);
1277 return err;
1278}
1279EXPORT_SYMBOL(xfrm_policy_register_afinfo);
1280
1281int xfrm_policy_unregister_afinfo(struct xfrm_policy_afinfo *afinfo)
1282{
1283 int err = 0;
1284 if (unlikely(afinfo == NULL))
1285 return -EINVAL;
1286 if (unlikely(afinfo->family >= NPROTO))
1287 return -EAFNOSUPPORT;
1288 write_lock(&xfrm_policy_afinfo_lock);
1289 if (likely(xfrm_policy_afinfo[afinfo->family] != NULL)) {
1290 if (unlikely(xfrm_policy_afinfo[afinfo->family] != afinfo))
1291 err = -EINVAL;
1292 else {
1293 struct dst_ops *dst_ops = afinfo->dst_ops;
1294 xfrm_policy_afinfo[afinfo->family] = NULL;
1295 dst_ops->kmem_cachep = NULL;
1296 dst_ops->check = NULL;
1da177e4
LT
1297 dst_ops->negative_advice = NULL;
1298 dst_ops->link_failure = NULL;
1da177e4
LT
1299 afinfo->garbage_collect = NULL;
1300 }
1301 }
1302 write_unlock(&xfrm_policy_afinfo_lock);
1303 return err;
1304}
1305EXPORT_SYMBOL(xfrm_policy_unregister_afinfo);
1306
1307static struct xfrm_policy_afinfo *xfrm_policy_get_afinfo(unsigned short family)
1308{
1309 struct xfrm_policy_afinfo *afinfo;
1310 if (unlikely(family >= NPROTO))
1311 return NULL;
1312 read_lock(&xfrm_policy_afinfo_lock);
1313 afinfo = xfrm_policy_afinfo[family];
1314 if (likely(afinfo != NULL))
1315 read_lock(&afinfo->lock);
1316 read_unlock(&xfrm_policy_afinfo_lock);
1317 return afinfo;
1318}
1319
1320static void xfrm_policy_put_afinfo(struct xfrm_policy_afinfo *afinfo)
1321{
1322 if (unlikely(afinfo == NULL))
1323 return;
1324 read_unlock(&afinfo->lock);
1325}
1326
1327static int xfrm_dev_event(struct notifier_block *this, unsigned long event, void *ptr)
1328{
1329 switch (event) {
1330 case NETDEV_DOWN:
1331 xfrm_flush_bundles();
1332 }
1333 return NOTIFY_DONE;
1334}
1335
1336static struct notifier_block xfrm_dev_notifier = {
1337 xfrm_dev_event,
1338 NULL,
1339 0
1340};
1341
1342static void __init xfrm_policy_init(void)
1343{
1344 xfrm_dst_cache = kmem_cache_create("xfrm_dst_cache",
1345 sizeof(struct xfrm_dst),
1346 0, SLAB_HWCACHE_ALIGN,
1347 NULL, NULL);
1348 if (!xfrm_dst_cache)
1349 panic("XFRM: failed to allocate xfrm_dst_cache\n");
1350
1351 INIT_WORK(&xfrm_policy_gc_work, xfrm_policy_gc_task, NULL);
1352 register_netdevice_notifier(&xfrm_dev_notifier);
1353}
1354
1355void __init xfrm_init(void)
1356{
1357 xfrm_state_init();
1358 xfrm_policy_init();
1359 xfrm_input_init();
1360}
1361
This page took 0.17171 seconds and 5 git commands to generate.